J. Math. Anal. Appl.

338 (2008) 292–298

www.elsevier.com/locate/jmaa

Bounded linear non-absolutely summing operators

D. Puglisi a , J.B. Seoane-Sepúlveda b,∗,1
a Department of Mathematical Sciences, Kent State University, Kent, OH 44242, USA
b Facultad de Ciencias Matemáticas, Departamento de Análisis Matemático, Universidad Complutense de Madrid, Plaza de las Ciencias 3,
28040 Madrid, Spain
Received 30 November 2006
Available online 23 May 2007
Submitted by G. Corach

Abstract
We show that, in certain situations, we have lineability in the set of bounded linear and non-absolutely summing operators.
Examples on lineability of the set Πp (E, F ) \ Ip (E, F ) are also presented and some open questions are proposed.
© 2007 Elsevier Inc. All rights reserved.

Keywords: p-Summing operators; p-Integral operators; Lineability

0. Preliminaries and background

Let E be a Banach space. A subset M of E is said to be lineable [1,5,9] if there exists an infinite-dimensional vector
space, V ⊂ M ∪ {0}. Following [7, p. 55] we say that, given two Banach
 spaces E and F , an operator T ∈ L(E, F ) is
absolutely summing if for each unconditionally convergent series ∞ x
i=1 i in E, the series ∞
i=1 T (xi ) is absolutely
convergent in F . As usual, we denote by Π(E, F ) to the space of absolutely summing operators from E to F (also
called 1-summing). In a more general definition, take any 1  p  ∞ and u ∈ L(E, F ). We say that u is p-summing
B ∗
if there is a probability measure μ and bounded linear operators a : Lp (μ) → ∞F and b : E → L∞ (μ) giving rise to
the commutative diagram

u jF B ∗
E F ∞F
b a

L∞ (μ) Lp (μ)
ip

* Corresponding author.
E-mail addresses: puglisi@math.kent.edu (D. Puglisi), jseoane@mat.ucm.es (J.B. Seoane-Sepúlveda).
1 Supported by MTM 2006-03531.

0022-247X/$ – see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmaa.2007.05.029
 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298 293

B ∗
where ip : L∞ (μ) → Lp (μ) is the formal identity, and jF : F → ∞F is the canonical isometric embedding. We
denote
 
πp (u) = inf a · b ,
and this infimum is extended over all measures μ and operators a, b as above. Πp (E, F ) denotes the Banach space of
all p-summing operators from E to F , which is a linear subspace of L(E, F ). Also, πp defines a norm in Πp (E, F )
with u  πp (u) for every u ∈ Πp (E, F ). In a similar manner, we shall say that a linear mapping u : E → F between
Banach spaces is a p-integral operator (1  p  ∞) if there is a probability measure μ and bounded linear operators
a : Lp (μ) → F ∗∗ and b : E → L∞ (μ) giving rise to the commutative diagram
u kF
E F F ∗∗
b a

L∞ (μ) Lp (μ)
ip

where ip : L∞ (μ) → Lp (μ) is the formal identity, and kF : F → F ∗∗ is the canonical isometric embedding. The
Banach space of all p-integral operators from E to F is denoted Ip (E, F ). Analogously, for each u ∈ Ip (E, F ) we
associate its p-integral norm,
 
ιp (u) = inf a · b ,
and this infimum is extended over all measures μ and operators a, b as above. The interested reader can refer to [4]
for a complete study of these classes of operators.

Definition 0.1. ABanach space E issaid to have the “two series property” provided there exist unconditionally
convergent series ∞ ∗
i=1 fi in E and
∞
i=1 xi in E such that
 ∞
∞ 
1
 |fj (xi )|2 2
= +∞.
fj 
i=1 j =1

Definition 0.2. A Banach space E is said to be “sufficiently Euclidean” if there exist a positive constant C and
sequences of operators {Jn }n∈N , {Pn }n∈N such that
J P
n2 −→
n
E −→
n
n2
with Pn ◦ Jn = In (the identity operator on n2 ) and Pn  = 1, 1  Jn   C, for every n ∈ N.

Stegall and Retherford [11] proved that every sufficiently Euclidean Banach space E has the two series property,
and every Lp -space (1 < p < ∞) is sufficiently Euclidean.
This note is divided in 2 main sections. In Section 1 we show that the set L(E, F ∗ ) \ Π1 (E, F ∗ ) is lineable,
where E is any Banach space with the two series property. Section 2 shows that the set Πp (E, F ) \ Ip (E, F ) is also
lineable for every p  1 (p = 2). We also propose some problems and give some directions to study the lineability of
certain sets of operators. Some results due to Dvoretzky and Pełczyński, some summability techniques, and some set
theoretical considerations are used.

1. The set L(E, F ∗ ) \ Π1 (E, F ∗ )

We start this section by introducing a basic lemma about divergent series.

∞
Lemma 1.1. Let {an }n∈N be a sequence of positive real numbers. If n=1 an = +∞ then there exists {Ai }i∈N ⊆ N
such that:
294 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298

(i) |Ai | = ω0 for each i ∈ N, where |A| denotes the cardinality of A, and ω0 = |N|,
Ai ∩ Aj = ∅ if i = j , and
(ii) 
(iii) m∈Ai am = +∞ for each i ∈ N.
∞
Proof. Since n=1 an = +∞, and an  0, ∀n ∈ N, by Cauchy’s criterion there exists ε > 0 such that
∀ν ∈ N ∃p ∈ N: aν + · · · + aν+p  ε.
Then, fixed ν1 ∈ N there is p1 ∈ N with aν1 + · · · + aν1 +p1  ε. Now, for each k ∈ N, consider νk ∈ N with νk 
νk−1 + pk−1 . Then there exists pk ∈ N such that aνk + · · · + aνk +pk  ε. Next, let

B1 = {ν2k , . . . , ν2k+p2k }
k∈N

and

A1 = {ν2k+1 , . . . , ν2k+1+p2k+1 }.
k∈N
 
We now have: B1 , A1 ⊆ N with |B1 | = |A1 | = ω0 , B1 ∩ A1 = ∅ and n∈B1 an = n∈A1 an = +∞.
Working
 on B
1 and for the same reason as above we have two countably infinite disjoint subsets B2 , A2 of B1 such
that n∈B2 an , n∈A2 an = +∞. Following this inductive procedure we obtain the desired sequence, {Ai }i∈N . 2
 ∞
Lemma 1.2. Let E be a Banach space enjoying the two series property. Let ∞ ∗
i=1 fi in E and i=1 xi in E such
∞ ∞ |fj (xi )|2 1
that i=1 [ j =1 fj  ] 2 = +∞. Then, there exists {An }n∈N ⊆ N such that:

(i) |An | = ω0 for each n ∈ N,

(ii) An ∩ Am = ∅ if n = m, and
∞  |fj (xi )|2 1
(iii) i=1 [ j ∈An fj  ] = +∞ for each n ∈ N.
2

|f (x )|
Proof. To simplify, let us denote by ai,j := fj 1/2 i
for each i, j ∈ N. Now, notice that one can assume that ai =
j
(ai,j )j ∈N ∈ 2 for each i ∈ N (otherwise the conclusion follows, directly, from the previous lemma). Thus, for some
positive elements ri = (ri,j )j ∈N of the unit sphere of 2 , we have

∞
 ∞ 1 ∞
 ∞ 1 ∞ ∞ ∞
 ∞  ∞
 ∞ 
  |fj (xi )|2 2   2      
= 2
ai,j = ai 2 = ai , ri  = ai,j ri,j = ai,j ri,j .
fj 
i=1 j =1 i=1 j =1 i=1 i=1 i=1 j =1 j =1 i=1

Next, let
∞

I := j ∈ N: ai,j ri,j = +∞ .
i=1

We need to consider two possible cases:

(1) I is infinite. In this case the assertion follows trivially.

(2) I is finite. In this
 case we can ∗suppose ∞ that I = ∅ since, considering permutations of index σ (i) and ρ(j ), we
can work with ∞ f
j =1 ρ(j ) in E and i=1 xσ (i) in E (still unconditionally convergent) in order to have I empty.
∞
Then, in this case, we can apply the previous lemma to the sequence sj = i=1 ai,j ri,j (note that ai,j ri,j  0 for
every i, j ∈ N). Thus, there exists {Ak }k∈N ⊆ N such that:
(i) |Ak | = ω0 for each i ∈ N,
Ak ∩ As
(ii)  = ∅ for k = s, and
∞
(iii) j ∈An i=1 ai,j ri,j = +∞ for each n ∈ N.
 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298 295

Finally,
∞
  ∞

+∞ = ai,j ri,j = ai,j ri,j  (ai,j )j ∈An 2
∀n ∈ N,
j ∈An i=1 i∈N j ∈An i=1

and we are done. 2

Theorem 1.3. Let E be a Banach space with the two series property. Then
L(E, 2 ) \ Π1 (E, 2 )
is lineable.
∞ ∞
Proof. By hypothesis, we know that there exist unconditionally convergent series i=1 fi in E ∗ and i=1 xi in E
such that
1
∞
 ∞
 |fj (xi )|2 2
= +∞.
fj 
i=1 j =1

Consider the sequence {An }n∈N as in the previous lemma.

For each n ∈ N, let us define
Tn : E −→ 2
by
 fk (x)
Tn (x) = 1
ek .
k∈An fk  2
 ∞
Since ∞ ∗
j =1 fj is unconditionally convergent in E , we have that supx=1 j =1 |fj (x)| is finite, call it C. Moreover,
for each x ∈ BX (the closed unit ball of X)
 fk (x)  1  1
|fk (x)|2 2 |fk (x)|2 2 1
Tn (x)  = e
1 k
=  C2.
f  f 
k∈An fk  2
2
2 k∈An k k
k∈N

This means that Tn is well-defined and Tn ∈ L(E, 2 ) for every n ∈ N.

Also, for every n ∈ N
∞ ∞   1
|fk (xi )|2 2
Tn (xi ) = = +∞
fk 
i=1 i=1 k∈An

from the choice of the sequence {An }n∈N . Since ∞ i=1 xi is unconditionally convergent in E, and by definition of
absolutely summing operator, we have that Tn ∈ / Π1 (E, 2 ) for each n ∈ N.
Now, from (ii) of the previous lemma, and from the definition, we also have that the sequence {Tn }n∈N is linearly
independent in L(E, 2 ).
Now, we have to show that every bounded linear operator belonging to the linear span of (Tn )n is not absolutely
summing. For simplicity, we consider the linear combination of two elements (the general case follows similarly).
Thus, let λ1 , λ2 ∈ K and n1 , n2 ∈ N, by definition
 fk (λ1 x)  fk (λ2 x)
λ1 Tn1 (x) + λ2 Tn2 (x) = 1
ek + 1
ek .
k∈An fk  2 1 k∈An fk  2 2

We can also assume, without loss of generality, that λ1 = 0. Then

∞     ∞    |fk ( λλ2 xi )|2  2 
1 ∞   1
xi xi |fk (xi )|2 |fk (xi )|2 2
λ1 Tn1 + λ2 Tn2 = + 1

λ1 λ 1 2 fk  fk  fk 
i=1 i=1 k∈An1 k∈An2 i=1 k∈An1
= +∞.
296 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298

∞ xi
Since i=1 λ1 is an unconditional convergent series in E we obtain that
span{Tn : n ∈ N} ⊆ L(E, F ) \ Π1 (E, F ) ∪ {0},
and we are done. 2

Let us recall a couple of well-known results, that we will need later [2,8].

Theorem 1.4 (Dvoretzky). For each ε > 0 and each n ∈ N, there exists n(ε) ∈ N such that if E is a Banach space of
dimension greater that n(ε), then there exists a subspace F of E with d(F, n2 )  1 + ε.

Proposition 1.5. Let X, Y be any two Banach spaces, and T : X → Y a bounded linear operator. Then there exists
a constant C so that T ∈ Π1 (X, Y ) and π1 (T )  1 if and only if for each finite subspace X0 of X the restriction
T |X0 ∈ Π1 (X0 , Y ) and π(T |X0 )  C.

Moreover, we know that, for any two Banach spaces E, F , we have L(E, F ∗ ) ∼ = L(F, E ∗ ), and Π1 (E, F ∗ ) ∼
=
Π1 (F, E ∗ ) (here ∼
= denotes an isometric isomorphism).
Thus, using the proposition above and Dvoretzky’s result, if T ∈ L(E, 2 ) but T is not absolutely summing, then
T is not absolutely summing when seen as an operator in L(2 , E ∗ ). Again, by the above proposition there exists
/ Π1 (n2 , E ∗ ). Now, by Dvoretzky’s result, there exists F0 subspace of F and S0 : F0 → n2
n ∈ N such that T |n2 ∈
with S0   1, S0−1   1 + . Therefore T ◦ S0 ∈ / Π1 (F0 , E ∗ ) (otherwise, and by the ideal property, we would
have T = S0−1 S0 T ∈ Π1 (n2 , E ∗ )). Finally, and using an argument about the projections, we can find T ∈ L(F, E ∗ )
and non-absolutely summing. Following in this way, given two Banach spaces E, F , with E enjoying the two series
property, we can construct a sequence Tn ∈ L(E, F ∗ ) non-absolutely summing from the sequence Tn ∈ L(E, 2 ) made
in Theorem 1.3. Finally, notice that Tn preserves the linear independence (because in our case (Tn )n ⊆ L(2 , E ∗ ) have
disjoint support, and to construct the sequence (Tn )n we used bijections and projections).
As a consequence we have

Corollary 1.6. Let E, F be two Banach spaces, with E having the two series property. Then
   
L E, F ∗ \ Π1 E, F ∗
is lineable.

2. The set Πp (E, F ) \ Ip (E, F ): An example

We will need the following lemma in this section. Its proof is well known.

Lemma 2.1. There exists a family F of subsets of N with:

(1) |F| = c, where c = |R|,

(2) for every Λ ∈ F , |Λ| = ω0 , and
(3) if Λ = Λ ∈ F , then Λ ∩ Λ is finite.

By means of the previous lemma we will construct a basis of an infinite-dimensional vector space every non-zero
element of which belongs to Πp (E, F ) \ Ip (E, F ), p  1, p = 2. We start by letting F = {Λα : α ∈ I } be a family as
in the previous lemma. Let us also fix N ∈ N and consider, for every α ∈ I , Uα ⊂ Λα such that |Uα | = N . It is well
known that in C(Uα ) the elements
2πkn
en (k) = e N

are a basis of C(Uα ) for 1  n  N . Then, each f ∈ C(Uα ) can be written, uniquely, as

f= fˆ(n)en .
n
 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298 297

Next, consider the measure μ, on Uα , given by μ({k}) = 1

N for every k ∈ Uα . Now, for each Aα ⊂ Uα let us define the
following subspaces:
  
CAα = f ∈ C(Uα ): f = fˆ(n)en ,
n∈Aα

and
  
Lp,Aα = f ∈ Lp (μ): f = ˆ
f (n)en .
n∈Aα

Also, denote by uAα the natural inclusion CAα → Lp,Aα . Clearly, πp (uAα )  1. Now, we want to estimate the value
of ιp (uAα ). We will follow a similar construction as in [4, pp. 103–104], sketching only the parts we need to change.
In order to do that, consider the p-integral factorization (see [4]) given by
i jp,α ũ
CAα −→
α
C(Uα ) −−→ Lp (ν) −→
α
Lp,Aα
where ν is a probability measure on Uα . Since |Λα | = ω0 , there is a bijection Iα : Λα ↔ N. Without loss of generality
we can assume that Iα (Uα ) = {1, 2, . . . , N}. For every f ∈ C(Uα ), k ∈ Z, define
  
fk (n) = f Iα−1 Iα (n) + Iα (k) , n ∈ Uα , (1)
where Iα (n) + Iα (k) is taken modulo N . Let us now define the following measure νk , given by
νk , f  = ν, fk .
N
It is easy to see that μ = 1
N · k=1 νk . Next, consider wα : C(Uα ) → Lp,Aα defined as

1 
N
wα (f ) = · (ũα jp,α fk )−k . (2)
N
k=1
Now, using the same argument as the ones from [4], one can obtain that
| p1 − 12 |
sup ιp (uAα )  Kp · N .
Aα
 
Defining now the spaces X = ( N CAα )2 and Y = ( N Lp,Aα )2 , and denoting
P uA J
ũαN : X −−→
N
CAα −−→
α
Lp (Aα ) −−→
N
Y,
N →∞
we obtain that πp (ũαN )  1 for all N ∈ N, and ιp (ũαN ) −−−−→ ∞. From the fact that Πp (X, Y ) is a dual space (see,
e.g. [10]) and using Cantor’s diagonalization process, we can find (nk )k ⊂ N such that
ũα = ω∗ − lim ũαnk
k

is well-defined. Thus, we obtain that ũα is p-summing but not p-integral. Also, and since the intersection of the sup-
ports of any finite number of these operators ũα is finite, we obtain that the family {ũα : α ∈ I } is linearly independent.
Now, it is not difficult to check that the family {ũα : α ∈ I } spans a vector space in (Πp (X, Y ) \ Ip (X, Y )) ∪ {0}.
Indeed, let λ1 , λ2 ∈ K and α1 , α2 ∈ I (as we did earlier, we will consider the linear combination of just two elements,
but, as the reader can check, the general case follows similarly). By the above construction,
λ1 ũαN1 + λ2 ũαN2 = JN ◦ (λ1 uAα1 + λ2 uAα2 ) ◦ PN
where PN and JN are the natural projection and embedding, respectively, and λ1 uAα1 + λ2 uAα2 : CAα1 ×Aα2 →
Lp (Aα1 × Aα2 ) is “almost” the natural map; the reader can see that we also obtain that
  | 1 −1|
sup {λ1 uAα1 + λ2 uAα2 }  Kp |λ1 |, |λ2 | N p 2
Aα1 ×Aα2

where the constant depends on |λ1 | and |λ2 |. Thus, λ1 ũα1 + λ2 ũα2 is not p-integral. We can summarize the previous
construction as follows:
298 D. Puglisi, J.B. Seoane-Sepúlveda / J. Math. Anal. Appl. 338 (2008) 292–298

Example 2.2. Let X and Y be the Banach spaces constructed above. Then Πp (X, Y ) \ Ip (X, Y ) is lineable for every
1  p < ∞, p = 2. Moreover, there exists a vector space V ⊂ (Πp (X, Y ) \ Ip (X, Y )) ∪ {0} such that dim(V ) = c.

The following remark is also worth mentioning.

Remark 2.3. In the previous example we need p = 2 since Π2 (X, Y ) = I2 (X, Y ) for every pair of Banach spaces
X, Y (see [4, Corollary 5.9]).

We finish by proposing some problems related to the lineability of certain subsets of operators in Banach spaces.

Question 2.4. In [3] Davis and Johnson proved that there exists a bounded linear operator that is not absolutely
summing under certain conditions, namely the set L(E, F ) \ Πp (E, F ) is non-empty whenever E is superreflexive
and F is any Banach space. One could ask whether this set is lineable. We believe that this is the case.

Question 2.5. In [6] Figiel and Johnson proved that, if a Banach space X has the approximation property but fails the
bounded approximation property, and X ∗ is separable, then there exists a non-nuclear operator T on X such that T ∗
is nuclear. Therefore, is the set of non-nuclear operators (and with nuclear adjoint) lineable?

Acknowledgments

The first author acknowledges the hospitality he received from the Departamento de Análisis Matemático at Universidad Complutense de
Madrid (Spain) during his visit there, while this paper was being written. The authors also express their gratitude to the referee whose insightful
remarks and comments improved the text.

References

[1] R.M. Aron, V.I. Gurariy, J.B. Seoane-Sepúlveda, Lineability and spaceability of sets of functions on R, Proc. Amer. Math. Soc. 133 (2005)
795–803.
[2] A. Dvoretzky, Some results on convex bodies and Banach spaces, in: Proc. Internat. Sympos. Linear Spaces, Jerusalem, 1960, Jerusalem
Academic Press, 1960, pp. 123–160.
[3] W.J. Davis, W.B. Johnson, Compact, non-nuclear operators, Studia Math. 51 (1974) 81–85.
[4] J. Diestel, H. Jarchow, A. Tonge, Absolutely Summing Operators, Cambridge University Press, 1995.
[5] P. Enflo, V.I. Gurariy, On lineability and spaceability of sets in function spaces, preprint.
[6] T. Figiel, W.B. Johnson, The approximation property does not imply the bounded approximation property, Proc. Amer. Math. Soc. 41 (1973)
197–200.
[7] A. Grothendieck, Produits tensoriels topologiques et espaces nucléaries, Mem. Amer. Math. Soc. 16 (1955).
[8] R.C. James, Superreflexive spaces, Canad. J. Math. 24 (1972) 896–904.
[9] F.J. García-Pacheco, N. Palmberg, J.B. Seoane-Sepúlveda, Lineability and algebrability of pathological phenomena in analysis, J. Math. Anal.
Appl. 326 (2007) 929–939.
[10] R. Ryan, Introduction to Tensor Products of Banach Spaces, Springer Monogr. Math., Springer-Verlag London, Ltd., London, 2002.
[11] C.P. Stegall, J.R. Retherford, Fully nuclear and completely nuclear operators with applications to L1 - and L∞ -spaces, Trans. Amer. Math.
Soc. 163 (1972) 457–492.